Visual Pattern-Driven Exploration of Big Data.

Area of Science:

• Data Science
• Computer Vision
• Bioinformatics

Background:

• Increasing data volumes generate numerous patterns, overwhelming analysts.
• Existing visualization methods struggle with overview questions on pattern quality and relevance.
• Effective exploration of large pattern result spaces remains a challenge.

Purpose of the Study:

• To develop a visual analytics pipeline for semi-automatic exploration of pattern result spaces.
• To address the challenge of managing and understanding vast numbers of extracted patterns.
• To improve the overview and detail analysis of pattern distributions, quality, and relevance.

Main Methods:

• Combined image feature analysis and unsupervised learning to partition pattern spaces.
• Developed novel quality metrics based on distance distributions for feature selection.
• Implemented an interactive visualization for hierarchical exploration from overview to detail.

Main Results:

• Successfully partitioned pattern spaces into interpretable and coherent chunks.
• Demonstrated effective guidance of feature selection using novel quality metrics.
• Showcased interactive drill-down capabilities from overview to detailed pattern analysis.

Conclusions:

• The proposed visual analytics pipeline facilitates efficient exploration of large pattern result spaces.
• The approach aids in prioritizing patterns for in-depth analysis without ground-truth.
• Validated through case studies in Earth observation and biomedical genomics data analysis.